On Wed, Aug 26, 2015 at 10:54:47AM +1000, Dave Chinner wrote:
> On Tue, Aug 25, 2015 at 05:33:05PM -0700, Darrick J. Wong wrote:
> > If during prefetch we encounter a bmbt block that fails the CRC check
> > due to corruption in the unused part of the block, force the buffer
> > back through the non-prefetch verifiers later so that the CRC is
> > updated. Otherwise, the bad checksum goes unfixed and the kernel will
> > still flag the bmbt block as invalid.
> >
> > Signed-off-by: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
> > ---
> > repair/prefetch.c | 8 ++++++++
> > 1 file changed, 8 insertions(+)
> >
> >
> > diff --git a/repair/prefetch.c b/repair/prefetch.c
> > index 1de3ec0..77d29c8 100644
> > --- a/repair/prefetch.c
> > +++ b/repair/prefetch.c
> > @@ -276,6 +276,14 @@ pf_scan_lbtree(
> >
> > XFS_BUF_SET_PRIORITY(bp, isadir ? B_DIR_BMAP : B_BMAP);
> >
> > + /*
> > + * Make this bmbt buffer go back through the verifiers later so that
> > + * we correct checksum errors stemming from bitflips in the unused
> > + * parts of the bmbt block.
> > + */
> > + if (bp->b_error == -EFSBADCRC)
> > + bp->b_flags |= LIBXFS_B_UNCHECKED;
>
> This is because the next read of the buffer clears bp->b_error,
> right?
>
> So, while I think this is necessary, I also think the prefetch on
> this btree should stop as we can't trust the contents of the buffer
> to be correct. Hence I'd suggest that:
>
> /*
> * If the verfier flagged a problem with the buffer, we
> * can't trust it's contents for the purposes of readahead.
> * Stop prefetching the tree, and mark this buffer as
> * unchecked so that the next read of the buffer by the
> * repair code will retain the error status and hence be
> * acted on appropriately.
> */
> if (bp->b_error) {
> bp->b_flags |= LIBXFS_B_UNCHECKED;
> libxfs_putbuf(bp);
> return 0;
> }
Ok, I'll change it to stop prefetch on the bmapbt.
--D
>
> Cheers,
>
> Dave.
>
>
> --
> Dave Chinner
> david@xxxxxxxxxxxxx
>
> _______________________________________________
> xfs mailing list
> xfs@xxxxxxxxxxx
> http://oss.sgi.com/mailman/listinfo/xfs
|